The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it interpolates conversation content and agent output directly into HTML templates. If the agent processes untrusted external data (such as code for review), that data could influence the report content to include malicious scripts or HTML. \n- Ingestion points: Conversation history and agent output ingested in SKILL.md. \n- Boundary markers: Absent; instructions require agent to reproduce content verbatim. \n- Capability inventory: Capability to write files to the local filesystem based on paths in output-config.toml. \n- Sanitization: Absent; no HTML escaping or filtering is applied to the content. \n- [DATA_EXFILTRATION]: The configuration file output-config.toml contains a hardcoded absolute path (C:/Users/aksh0/reports), which reveals a specific directory structure on the local host. While this is a data exposure concern regarding the author's environment, it does not involve unauthorized transmission of data to a remote source.

save-report